Two-Stage Deformulation Sampling for Ingredient Amount Estimation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for reverse engineering ingredient amounts in consumer goods are inefficient and inaccurate, particularly in the cosmetics industry, where the exact amounts of ingredients are unknown, leading to waste and low accuracy.

Innovation Solution

A two-stage machine learning approach using variational auto-encoders (VAEs) and sampling techniques to efficiently determine ingredient amounts by leveraging the structure within the VAE latent space, reducing computational resources and improving processing power.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If single-stage sampling methods are used for deformulation, then the process is simpler to implement, but computational costs are high and accuracy is low

Engineering Contradiction:
Improveease of implementationVSAvoidaccuracy of ingredient amount determination
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The deformulation process is divided into two distinct stages: (1) generating an initial set of candidate formulations using random sampling, and (2) performing targeted sampling within the latent space to refine these candidates. This segmentation allows the system to first explore the solution space broadly, then focus computational resources on promising regions, thereby improving accuracy without excessive complexity

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The first stage performs preliminary sampling to identify promising candidate formulations before committing to detailed analysis. By pre-filtering candidates based on initial sampling results, the system eliminates obviously poor formulations early, reducing overall computational cost while maintaining high accuracy in the final results

Inventive Principle:
Principle #10Preliminary action

2Ease of manufacture

If single-stage sampling methods are used for deformulation, then the implementation is simpler, but computational resources are wasted

Engineering Contradiction:
Improveease of implementationVSAvoidcomputational resource waste
Core Design Contradiction:
Ease of manufactureVSLoss of energy

Solution Approach 1:

The two-stage approach segments the computational workload: Stage 1 performs broad but efficient random sampling to identify promising regions, while Stage 2 concentrates computational resources on refined sampling within the latent space of these promising candidates. This prevents wasting resources uniformly across the entire solution space and reduces overall computational energy consumption

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Rather than performing exhaustive sampling across the entire formulation space, the method applies partial action by focusing sampling efforts only on promising regions identified in Stage 1. This selective approach significantly reduces computational resource waste while still achieving accurate deformulation results

Inventive Principle:
Principle #16Partial or excessive action

3Ease of operation

If existing non-chemical deformulation techniques are used, then chemical expertise is not required, but the results produce waste and have low accuracy

Engineering Contradiction:
Improveno chemical expertise requiredVSAvoidaccuracy of ingredient amount determination
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent replaces traditional chemical analysis methods with a machine learning-based computational approach using variational autoencoders. This substitution maintains the advantage of not requiring chemical expertise while dramatically improving accuracy by leveraging patterns in formulation data to predict ingredient amounts from ordered lists

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

4Reliability

If exhaustive sampling is performed in the original formulation vector space, then all possible formulations are explored, but the computational cost is prohibitively high

Engineering Contradiction:
Improvecompleteness of formulation explorationVSAvoidcomputational cost
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The method transforms the sampling problem from the high-dimensional original formulation space to a compressed latent space created by the variational autoencoder. This dimensional transformation allows the system to explore formulation variations more efficiently by sampling in the lower-dimensional latent space, then mapping results back to the original space, thereby maintaining reliability while reducing computational energy use

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4285371B1Two-stage sampling for accelerated deformulation generation
Publication Date: 2026.01.14 CITRINE INFORMATICS INC
  • EP4285371B1 patent drawingFigure 1
  • EP4285371B1 patent drawingFigure 2
  • EP4285371B1 patent drawingFigure 3

AI summary

A device receives an ingredient list having a sequence of ingredients ordered by relative amount, and generates formulation vectors by sampling the ingredients list. The device inputs the plurality of formulation vectors into a machine-learned model, the machine-learned model generating an encoded version of each of the plurality of formulation vectors using an encoder, and then outputting a plurality of reconstructed formulation vectors as derived using a decoder. The device identifies reconstructed formulation vectors that have an order that matches the sequence, defines a latent space using the encoded version of the matching reconstructed formulation vectors. The device iteratively samples the latent space until a threshold number of samples are derived that match an ordering constraint that corresponds to the sequence, performs a statistical aggregation of the samples, and outputs an indication of an absolute amount of each ingredient in the ingredients list.